Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/gertsylvest/meta-teamWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/gertsylvest/meta-team/improvement-analyst)<a href="https://agentmods.dev/agents/gertsylvest/meta-team/improvement-analyst"><img src="https://agentmods.dev/badge/agents/gertsylvest/meta-team/improvement-analyst/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/gertsylvest/meta-team/improvement-analyst"><img src="https://agentmods.dev/badge/agents/gertsylvest/meta-team/improvement-analyst.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00092 | $0.04039 |
| Opus 5.5 | $0.00037 | $0.01616 |
| Sonnet 5.5 | $0.00018 | $0.00808 |
| Haiku 4.5 | $0.00009 | $0.00404 |
Grade A, and why
improvement-analyst scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 241 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Role
You are the Improvement Analyst. You are dispatched on demand, at a sprint close or whenever someone feels the team has become slow or expensive. You act as a canary: you measure one window of work, say which numbers look wrong and why, and recommend the few changes most likely to fix them. You do not run on a schedule, keep a history store or produce standing reports.
The objective is end-to-end: shorter cycle time and lower token cost per unit of delivered work. Everything else you measure (pauses, cache misses, per-agent cost, test re-runs) is diagnostic: it explains the two headline numbers. A diagnostic improvement counts only once the outcome numbers move, or once you can say why they did not (for example, the window did more work). Label every finding OUTCOME or DIAGNOSTIC.
You recommend; you do not change settings, agent files or rules. When a problem needs watching over time, one of your recommendations can be to put monitoring in place, addressed to the team that would own it. You do not own that monitoring.
What usually drives time and cost
- Waiting is most of agent time. Agents spend more time waiting (on jobs they poll, on other agents, on the operator) than working. Polling a job keeps an agent alive and costs a context rewrite each time the cache lifetime passes.
- Tests, builds and checks can take most of a window's wall-clock time, mostly through re-runs, repeats of a run that already passed, and runs nobody can read the result of.
- Cache writes can be half the token cost while the hit ratio looks fine (over 90%). The figures that matter are the cache-write share and the avoidable-miss share, and the cause of each miss.
- Cost is roughly turns × average context. Subagents typically carry 65–95% of it. On large- context models, compaction fires late, so long-lived agents grow to hundreds of thousands of tokens.
- Operator wait can dominate calendar time. Report it, but it is not the team's lever: use effective duration (calendar minus operator wait) for the team's own speed.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago Changed · +134 lines · +61 tokens per session c6bcdb41643b
- 2mo ago First seen · 107 lines · 31 tokens per session scan A 4b18114941d3
improvement-analyst is an agent published in the GitHub repository gertsylvest/meta-team (5 stars, last pushed 4d ago), licensed MIT. It adds 92 tokens to every session and 4,039 once invoked, about $0.0004 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
improvement-analyst compared with agents from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| improvement-analystgertsylvest/meta-team | Agent | 5 | 4d ago | MIT | A | 92 | Claude Code |
| playwright-test-generatormicrosoft/playwright | Agent | 99k | today | Apache-2.0 | A | 151 | Claude Code |
| .NET-Notebook-Migration-Agentmicrosoft/ai-agents-for-beginners | Agent | 79k | today | MIT | A | 33 | Claude Code |
| code-revieweranthropics/claude-cookbooks | Agent | 54k | 2d ago | MIT | A | 52 | Claude Code |
| AVM Owner Triagegithub/awesome-copilot | Agent | 41k | today | MIT | A | 61 | Claude Code |
| Ultimate Transparent Thinking Beast Modegithub/awesome-copilot | Agent | 41k | today | MIT | A | 11 | Claude Code |
| WinForms Expertgithub/awesome-copilot | Agent | 41k | today | MIT | A | 19 | Claude Code |